Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)
Current Result Document : 21 / 21
ÀÌÀü°Ç
ÇѱÛÁ¦¸ñ(Korean Title) |
ÁÖÀÇÁýÁß ¹× º¹»ç ÀÛ¿ëÀ» °¡Áø Sequence-to-Sequence ¼øȯ½Å°æ¸ÁÀ» ÀÌ¿ëÇÑ Á¦¸ñ »ý¼º ¸ðµ¨ |
¿µ¹®Á¦¸ñ(English Title) |
Title Generation Model for which Sequence-to-Sequence RNNs with Attention and Copying Mechanisms are used |
ÀúÀÚ(Author) |
ÀÌÇö±¸
±èÇмö
Hyeon-gu Lee
Harksoo Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 44 NO. 07 PP. 0674 ~ 0679 (2017. 07) |
Çѱ۳»¿ë (Korean Abstract) |
´ë¿ë·®ÀÇ ÅؽºÆ® ¹®¼°¡ ¸ÅÀÏ ¸¸µé¾îÁö´Â ºòµ¥ÀÌÅÍ È¯°æ¿¡¼ Á¦¸ñÀº ¹®¼ÀÇ ÇÙ½É ¾ÆÀ̵ð¾î¸¦ ºü¸£°Ô Áý¾î³»´Âµ¥ ¸Å¿ì Áß¿äÇÑ ´Ü¼°¡ µÈ´Ù. ±×·¯³ª ºí·Î±× ±â»ç³ª ¼Ò¼È ¹Ìµð¾î ¸Þ½ÃÁö¿Í °°Àº ¸¹Àº Á¾·ùÀÇ ¹®¼µéÀº Á¦¸ñÀ» °®°í ÀÖÁö ¾Ê´Ù. º» ³í¹®¿¡¼´Â ÁÖÀÇÁýÁß ¹× º¹»ç ÀÛ¿ëÀ» °¡Áø sequence-to-sequence ¼øȯ½Å°æ¸ÁÀ» »ç¿ëÇÑ Á¦¸ñ »ý¼º ¸ðµ¨À» Á¦¾ÈÇÑ´Ù. Á¦¾È ¸ðµ¨Àº ¾ç¹æÇâ GRU(Gated Recurrent Unit) ³×Æ®¿öÅ©¿¡ ±â¹Ý ÇÏ¿© ÀÔ·Â ¹®ÀåÀ» ÀÎÄÚµù(encoding)ÇÏ°í, ÀÔ·Â ¹®Àå¿¡¼ ÀÚµ¿ ¼±º°µÈ Å°¿öµå¿Í ÇÔ²² ÀÎÄÚµùµÈ ¹®ÀåÀ» µðÄÚµùÇÔÀ¸·Î½á Á¦¸ñ ´Ü¾îµéÀ» »ý¼ºÇÑ´Ù. 93,631¹®¼ÀÇ ÇнÀ µ¥ÀÌÅÍ¿Í 500¹®¼ÀÇ Æò°¡ µ¥ÀÌÅ͸¦ °¡Áø ½ÇÇè¿¡¼ ÁÖÀÇÁýÁß ÀÛ¿ë¹æ¹ýÀÌ º¹»ç ÀÛ¿ë¹æ¹ýº¸´Ù ³ôÀº ¾îÈÖ ÀÏÄ¡À²(ROUGE-1: 0.1935, ROUGE-2: 0.0364, ROUGE-L: 0.1555)À» º¸¿´°í »ç¶÷ÀÌ Á¤¼ºÆò°¡ÇÑ ÁöÇ¥´Â º¹»ç ÀÛ¿ë¹æ¹ýÀÌ ³ôÀº ¼º´ÉÀ» º¸¿´´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
In big-data environments wherein large amounts of text documents are produced daily, titles are very important clues that enable a prompt catching of the key ideas in documents; however, titles are absent for numerous document types such as blog articles and social-media messages. In this paper, a title-generation model for which sequence-to-sequence RNNs with attention and copying mechanisms are employed is proposed. For the proposed model, input sentences are encoded based on bi-directional GRU (gated recurrent unit) networks, and the title words are generated through a decoding of the encoded sentences with keywords that are automatically selected from the input sentences. Regarding the experiments with 93631 training-data documents and 500 test-data documents, the attention-mechanism performances are more effective (ROUGE-1: 0.1935, ROUGE-2: 0.0364, ROUGE-L: 0.1555) than those of the copying mechanism; in addition, the qualitative-evaluation radiative performance of the former is higher.
|
Å°¿öµå(Keyword) |
sequence-to-sequence ¸ðµ¨
ÁÖÀÇÁýÁß ÀÛ¿ë
º¹»ç ÀÛ¿ë
Á¦¸ñ »ý¼º
¼øȯ½Å°æ¸Á
sequence-to-sequence model
attention mechanism
copying mechanism
title generation
recurrent neural network
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|